As regular readers will know, this blog carries monthly updates of Fulcrum’s activity “nowcasts” for all of the major global economies on a monthly basis. We believe that this is one of the few publicly available sources of regular updates of global nowcasts, which are rapidly becoming an important frame of reference for investors seeking to interpret the latest fluctuations in economic activity in the major economies.
In the US, for example, the Atlanta Fed’s GDPNow model is very widely followed, and the New York Fed has just started to publish weekly updates from a nowcast model on a weekly basis. The NY Fed nowcasting report may become the market’s preferred “benchmark” to assess the recent behaviour of US activity.
However, these two Fed models do not use the same methodology, and the term “nowcasting” can mean very different things to different economists. This has already caused confusion about the recent state of the US economy, which has slowed down substantially in Q1 according to the Atlanta Fed, and has strengthened only slightly in Q2, according to the NY Fed model. The Fulcrum model however suggests firmer growth during March and April.
In today’s blog (and in the attached paper, co-written with Juan Antolin-Diaz), we attempt to explain the key differences between the various approaches. Apologies in advance – this is a somewhat technical explanation, but these details are becoming important for investors and policy makers to understand.
The main distinction to be drawn is between:
- GDP tracking models, sometimes known as “bean counters”, like the Atlanta GDPNow model; and
- Dynamic Factor Models (DFMs) that identify the underlying activity growth rate in the economy from multiple economic data series, based on the pioneering econometric methodology of Evans (2005) and Giannone, Reichlin and Small (2008). Both the NY Fed and the Fulcrum models fall into this broad category.
Within the DFM group of models, there is a another distinction, in the presentation of the results:
- The NY Fed presents results that focus on GDP growth in specific quarters
- Fulcrum presents results in the FT and elsewhere that focus on the latest growth rate in underlying activity on a continuous basis, without attempting to forecast GDP specifically, and without focusing on calendar quarters.
Nowcasts are obviously intended to provide estimates of what is happening to activity now. We believe that the last category is the most appropriate for doing this. This output is especially useful for tracking financial asset prices, which are liable to shift by large amounts without reference to the noisy and backward-looking GDP data.
The latest results from Atlanta, the NY Fed and Fulcrum are summarised here:
The large differences in the latest growth rates estimated for the US economy are clear from the graph, which shows a variation in US growth from 0.3 per cent to 2.0 per cent, as of last Friday’s data. This is obviously too wide a variation to be useful to investors, at least without a full understanding of why such differences may exist. The main factors are as follows:
1. The Atlanta GDPNow model
This is a very sophisticated “bean counting” model which tries to mimic “key elements of the data construction machinery of the Bureau of Economic Analysis” to form “a relatively precise estimate of what the BEA will announce for the previous quarter’s GDP even before it is announced” (Faust and Wright, 2009). These models are typically accounting tools in which incoming monthly data relating to the components of GDP (consumption, investment, exports, etc.) are aggregated to produce an estimate of GDP itself.
This procedure has a good record of “predicting” the first BEA announcement of GDP, especially towards the end of the calendar quarter, and in the month between the quarter end and the “advance” GDP announcement. But it has the disadvantage that it becomes increasingly backward looking as the announcement gets closer.
At present, GDPNow reckons that the 2016Q1 GDP announcement will show a growth rate of only 0.3 per cent, when it is made by the BEA on 28 April. This very low growth rate has been widely cited, even though it is already very out-of-date. It is heavily influenced by economic data relating to January and February, when the economy was clearly depressed. The big question now is whether there has been a pick-up since then.
2. The New York Fed nowcast
The NY Fed nowcasts use modern Kalman filtering techniques and DFM models. Essentially, this method uses a large number of economic indicators, including many that are not specifically used to construct GDP (eg employment data and confidence surveys) to extract the “common factor” which captures the bulk of the business cycle fluctuations in the economy. This factor is then scaled to fit the pattern of typical GDP fluctuations.
The output from the NY model goes a step further. It uses these underlying activity factors to create GDP projections for specific calendar quarters. This is not likely to predict GDP in the most recent quarter as accurately as the Atlanta method, but it also predicts the next quarter, therefore giving us a better idea about ongoing GDP growth now and in the immediate future. At present, the NY estimate for Q2 is 1.2 per cent, well above the Atlanta estimate for Q1.
3. The Fulcrum nowcasts
The Fulcrum model is very similar to the NY Fed model, but it does not focus on GDP for specific quarters. Instead, it is calculated daily to estimate the continuous underlying growth rate in activity at any given point in time, with the objective of capturing changes in the business cycle as rapidly as possible.
The philosophy here is that GDP itself is not necessarily the “true” measure of underlying economic activity, but is subject to considerable measurement error and other forms of statistical noise. It is well known that GDP within any given calendar quarter can be a very poor indicator of underlying activity growth, as we have seen on many occasions in the current economic cycle.
The Fulcrum nowcast suggest that US activity growth fell continuously from the beginning of 2015 to February 2016, by which time it was around 1.0 per cent. However, in a potentially important change, the nowcast moved sharply higher in March and April, and it is now fluctuating around 2.0-2.5 per cent. This change was rapidly reflected in the prices of US risk assets, which recovered slightly before, and then along with, the daily US nowcasts.
Financial markets do not wait for quarterly GDP to be published, and they often ignore it altogether when it does finally appear. We prefer to ignore the noise from quarterly GDP, while focusing attention on the underlying activity factor that is driving the business cycle.
Investors are likely to track all of these nowcasts quite carefully in the future, and will come to understand why they may differ. This will relegate the quarterly roller coaster of GDP growth further into the sidelines. At present, the latest nowcasts indicate that US activity growth is much less bleak than the (likely) weak GDP release for 2016 Q1 will indicate.
Get alerts on Global Economy when a new story is published